net: tcp: fix unexcepted socket die when snd_wnd is 0
authorMenglong Dong <imagedong@tencent.com>
Fri, 11 Aug 2023 02:55:29 +0000 (10:55 +0800)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Wed, 13 Sep 2023 07:42:32 +0000 (09:42 +0200)
[ Upstream commit e89688e3e97868451a5d05b38a9d2633d6785cd4 ]

In tcp_retransmit_timer(), a window shrunk connection will be regarded
as timeout if 'tcp_jiffies32 - tp->rcv_tstamp > TCP_RTO_MAX'. This is not
right all the time.

The retransmits will become zero-window probes in tcp_retransmit_timer()
if the 'snd_wnd==0'. Therefore, the icsk->icsk_rto will come up to
TCP_RTO_MAX sooner or later.

However, the timer can be delayed and be triggered after 122877ms, not
TCP_RTO_MAX, as I tested.

Therefore, 'tcp_jiffies32 - tp->rcv_tstamp > TCP_RTO_MAX' is always true
once the RTO come up to TCP_RTO_MAX, and the socket will die.

Fix this by replacing the 'tcp_jiffies32' with '(u32)icsk->icsk_timeout',
which is exact the timestamp of the timeout.

However, "tp->rcv_tstamp" can restart from idle, then tp->rcv_tstamp
could already be a long time (minutes or hours) in the past even on the
first RTO. So we double check the timeout with the duration of the
retransmission.

Meanwhile, making "2 * TCP_RTO_MAX" as the timeout to avoid the socket
dying too soon.

Fixes: 1da177e4c3f4 ("Linux-2.6.12-rc2")
Link: https://lore.kernel.org/netdev/CADxym3YyMiO+zMD4zj03YPM3FBi-1LHi6gSD2XT8pyAMM096pg@mail.gmail.com/
Signed-off-by: Menglong Dong <imagedong@tencent.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Sasha Levin <sashal@kernel.org>
net/ipv4/tcp_timer.c

index cf354c2..44b49f7 100644 (file)
@@ -441,6 +441,22 @@ static void tcp_fastopen_synack_timer(struct sock *sk, struct request_sock *req)
                          req->timeout << req->num_timeout, TCP_RTO_MAX);
 }
 
+static bool tcp_rtx_probe0_timed_out(const struct sock *sk,
+                                    const struct sk_buff *skb)
+{
+       const struct tcp_sock *tp = tcp_sk(sk);
+       const int timeout = TCP_RTO_MAX * 2;
+       u32 rcv_delta, rtx_delta;
+
+       rcv_delta = inet_csk(sk)->icsk_timeout - tp->rcv_tstamp;
+       if (rcv_delta <= timeout)
+               return false;
+
+       rtx_delta = (u32)msecs_to_jiffies(tcp_time_stamp(tp) -
+                       (tp->retrans_stamp ?: tcp_skb_timestamp(skb)));
+
+       return rtx_delta > timeout;
+}
 
 /**
  *  tcp_retransmit_timer() - The TCP retransmit timeout handler
@@ -506,7 +522,7 @@ void tcp_retransmit_timer(struct sock *sk)
                                            tp->snd_una, tp->snd_nxt);
                }
 #endif
-               if (tcp_jiffies32 - tp->rcv_tstamp > TCP_RTO_MAX) {
+               if (tcp_rtx_probe0_timed_out(sk, skb)) {
                        tcp_write_err(sk);
                        goto out;
                }